Vulnerability of quantum classification to adversarial perturbations

Abstract

It is expected that the quantum advantage of certain classification problems, like quantum-enhanced machine learning algorithms, will become more pronounced as the Hilbert-space dimension grows, but care must be taken with perturbations on the system, as they lead to security breaches that can be explored by untrusted parties. Here it is shown that the amount of perturbation needed for an adversary to induce a misclassification scales inversely with the dimensionality of the space, but security can be restored if we restrict the class of states to be classified.